fix: report accurate is_running after trader update to stop redundant start

Live verification of the relaunch (update) branch surfaced this: the update
response omitted is_running, so the launcher fired a redundant start that got
400 'Trader is already running', showing an error toast and skipping the
dashboard navigation even though the relaunch succeeded.

- PUT /traders/:id now reports is_running (a running trader is restarted with
  the new config, so it stays running).
- The already-running rejection carries error_key trader.start.already_running
  and the shared launcher treats it as success (launch is idempotent).

api/handler_trader.go

@@ -731,7 +731,10 @@ func (s *Server) handleUpdateTrader(c *gin.Context) {
 		"trader_id":   traderID,
 		"trader_name": req.Name,
 		"ai_model":    req.AIModelID,
-		"message":     "Trader updated successfully",
+		// A running trader is restarted with the new config (async above), so
+		// report it as running — callers must not fire a redundant start.
+		"is_running": wasRunning,
+		"message":    "Trader updated successfully",
 	})
 }
 
@@ -792,7 +795,10 @@ func (s *Server) handleStartTrader(c *gin.Context) {
 	if existingTrader != nil {
 		status := existingTrader.GetStatus()
 		if isRunning, ok := status["is_running"].(bool); ok && isRunning {
-			c.JSON(http.StatusBadRequest, gin.H{"error": "Trader is already running"})
+			c.JSON(http.StatusBadRequest, gin.H{
+				"error":     "Trader is already running",
+				"error_key": "trader.start.already_running",
+			})
 			return
 		}
 		// Trader exists but is stopped - remove from memory to reload fresh config

web/src/lib/launch/launchAutopilot.test.ts

@@ -129,6 +129,33 @@ describe('launchAutopilot', () => {
     )
   })
 
+  it('treats a racing already-running rejection as success', async () => {
+    mocks.runLaunchPreflight.mockResolvedValue(readyPreflight())
+    mocks.api.getTraders.mockResolvedValue([
+      { trader_id: 't-1', trader_name: 'NOFX Autopilot', is_running: false },
+    ])
+    mocks.api.updateTrader.mockResolvedValue({
+      trader_id: 't-1',
+      is_running: false,
+    })
+    mocks.api.startTrader.mockRejectedValue(
+      new ApiError(
+        'Trader is already running',
+        'trader.start.already_running',
+        undefined,
+        400
+      )
+    )
+
+    const outcome = await launchAutopilot({
+      ensureStrategy: vi.fn().mockResolvedValue('strat-1'),
+    })
+
+    expect(outcome).toEqual(
+      expect.objectContaining({ ok: true, traderId: 't-1' })
+    )
+  })
+
   it('surfaces the server-side preflight result when start is rejected', async () => {
     mocks.runLaunchPreflight.mockResolvedValue(readyPreflight())
     mocks.api.startTrader.mockRejectedValue(

web/src/lib/launch/launchAutopilot.ts

@@ -101,7 +101,16 @@ export async function launchAutopilot(
       : await api.createTrader(traderRequest)
 
     if (!autopilot.is_running) {
-      await api.startTrader(autopilot.trader_id)
+      try {
+        await api.startTrader(autopilot.trader_id)
+      } catch (err) {
+        // Launch is idempotent: the update path restarts a running trader
+        // asynchronously, so a racing "already running" rejection is success.
+        const alreadyRunning =
+          err instanceof ApiError &&
+          err.errorKey === 'trader.start.already_running'
+        if (!alreadyRunning) throw err
+      }
     }
 
     return {